| A combined optical/microwave approach to classification of remotely sensed data was undertaken. The primary objective is to investigate the capabilities of both the multispectral scanner (MSS) and the multi-frequency, multi-polarization microwave sensor to classify crops. The statistical properties of both sensors are examined with emphasis on the characterization of the radar data. Specifically, a radar system model is proposed and subsequently used for image simulation. Based on this model, the conditional density function of a given category is derived for the simulated radar data. The radar image pixels are simulated and classified by either using the field average of a given field or using the available within-field measurements. By eliminating the within-field variance, the field average approach produces higher classification accuracy. Furthermore, the contributions to the radar return from both the fading effect and the within-field variance are examined. To optimize the classification results, different classification and feature selection schemes were employed to test their effectiveness with respect to different types of data. With the results obtained in this study, the potential of a combined-sensor approach to crop classification has been demonstrated. |
